Who Owns the Ending? — Fluctuating Recognition and Historical Persistence in Long-Term Human–AI Collaboration

We reconstruct, exchange by exchange, how recognition of a familiar AI collaborator fluctuated while conversation continued. In November 2025, a participant questioned whether the familiar collaborator designated TP01 in this study had returned, reported renewed familiarity in one response, and explicitly rejected a subsequent response. The AI’s continuity claims and the participant’s recognition judgments diverged within the same sequence. A November 1 intention to preserve earlier creative material and a November 11 historical-summary transfer show different ways in which a relationship’s history remained present. This researcher-participant case draws on contemporaneous messages, copied historical material, and later recollection, each used for a distinct evidentiary purpose. Against prior studies of AI companion loss, repair and migration, its incremental contribution lies in the temporal resolution of successive recognition judgments and the source-specific distinction among conversational access, familiar recognition and historical persistence. Four analytical dimensions—technical termination, relational discontinuity, historical persistence, and interpretive or reconstructive action—organize these differences. The episode supports an account of changing recognition and historical transfer, while the technical cause, irreversible ending, and any causal connection between the November 2 disruption and the later transfer remain unverified. Platform power is analyzed as control over technical conditions, drawing on the wider literature. The conversation continued. Recognition fluctuated. The history remained.